The NYC no wave/noise rock/post punk bands Sonic Youth and Swans are not only known for their eccentric and hypnotic noise landscapes but equally well for their intricate lyrics. While Sonic Youth’s lyrics are deeply rooted in the tradition-less tradition of modern American poetry, listening to Swans often is reminiscent of going to a church full of noise when singer Michael Gira recites his ecclesiastical texts in their maelstrom of noise. What both bands share is their love for repetition and noise.
Here, I will take a deep dive into the words of each of the two bands’ 15 studio records between 1983 and 2019 and provide different perspectives on some features of their lyrical craft. We’ll see some (uncanny) commonalities and defining differences. Some of the questions I try to answer here include:
If you’re interested in the code of this project, have a look here.
So, let’s get started.
I’ll get the lyrics from Genius’ API. You need a genius account and create an API client on https://genius.com/api-clients for this.
So, we got 4602 lines of poetry from 149 Sonic Youth song and 4407 lines of prayers from 146 Swans songs. Let’s see how many songs there are on each of the records.
Arrrrghrghrgrhgr! For reasons still unknown, we got lyrics for only 3/10 songs from “To Be Kind” and only 3/8 from “The Glowing Man”. But okej, we’ll take what we got and start spiraling in.
To get an idea of the lyrics we got from each record, let’s have a look at how many words they use on each record. Keep in mind that the lyrics of two Swans records didn’t load completely, thus, they’re a bit shorter than the rest.
From the last two plots, we see that both bands started with fewer songs on their first few records and - as success kicked in - the records got longer and so did the lyrics. Let’s look at how long their lyrics are per song.
Here’s a quick and dirty count of unique words per song. We exclude all the la la la’s and uh uh uh’s. The longest and shortest lyrics of each band are highlighted an labeled.
Doesn’t take (m)any words for a shaman to do their job. Brains i/o.
Repetitive lyrics always get me! Five unique words (because we did not stem strikes to strike, yet)! Though, admittedly, this closing song is clearly an ode to anti-art and rather a convulsion of guitar squeaks and spoken boredom.
The other end of this spectrum marks the song In the Kingdom #19, a gripping piece of poetry with dystopic guitars that seems to be a young descendant of Howl. 230 unique words, and, except for the chorus, quite the opposite of repetitiveness.
Now that we have some ballpark figures about how short or long their lyrics are, let’s have a look at what words they actually use (a lot, a lot).
Here are two word clouds of the words they use most often. Swans on the left, Sonic Youth on the right. Note that some words look a bit strange ( littl?! someth??!). That’s because the words were stemmed.
Time! Believe! Hey! And * drum rolls * Love! Music that sounds like a sledgehammer at times is no excuse to avoid the big L. It’s astonishing how different the two clouds look, pointing to their distinct vocabulary. Many of the words immediately trigger an Ohrwurm.
Now let’s have a look at how often they use frequent words they have in common and in how many songs they use them.
230 times love! In a mere 36 Swans songs. Sonic Youth doesn’t use it as often but in 49 songs. This trophy of repetitiveness goes to Swans! Sonic Youth, on the other hand, uses the word eye in almost half of their songs.
Below you see words that occur often alongside or close to each other. The darker the connecting line, the stronger their link. Swans on the left, Sonic Youth on the right.
A lot of Swans’ words cluster heavily together. In fact, I had to filter Swans’ lyrics quite a lot to make it easier to look at. Sonic Youth has a lot of “islands” word co-occurrences, showing that they use these words together but rarely with other words (e.g. walking and street, sun, start and coming).
So, coming back to the title of this little analysis, Repetition and Noise, how repetitive are their lyrics actually?
Here’s a plot of the two bands’ lexical variety of each of their records over time with the most and least repetitive record of both bands annotated.
Spännande! Both bands started VERY repetitive and then reached a plateau of absolute non-repetitiveness (a value close to one suggests no repetition!). Only Swans had a little detour into repetition-land after 2010 with their records “To be Kind” and “The Glowing Man” (but, note again, those are the records we didn’t get all the lyrics for).
Both bands had their first notable successes in the early 90s, after continually becoming less and less repetitive. Apparently, repetitive lyrics weren’t valued that much then.
A final urging question is if their lyrics are really as negative as I usually think they are. So, let’s have a look at the sentiment of their words.
Here’s the two bands’ sentiment (positive or negative) per record over time. Again, the most positive and negative records of both bands is annotated.
Well, as expected, Swans is extreme to both ends. The very negative “Cop” and the very positive “Holy Money” mark the extremes of this plot. Interestingly, both Swans and Sonic Youth started negative and over the years reached a plateau of almost neutral lyrics. Again, one could speculate whether somewhat positive lyrics were necessary for them to reach their first successes in the early 90s, or, vice versa, if success setting in calmed their troubled souls and they cheered up a bit (Note to self: This might be a nice little future project!)
However, if you know the Swans record “Holy Money”, you certainly protest now and type an angry email telling me that this is certainly not a positive record. I agree. This nicely illustrates the shortcomings of this type of sentiment analysis.
Love seems to be attached with a very positive notion and if the context is disregarded, you’re told this is a feel-good summer hit. Well…
Lastly, I wanted to know whether we can teach the computer to differentiate between the lyrics of Swans and Sonic Youth.
I trained several models and, as expected, they overfitted. Always. I tried everything from recursive partitioning over xgboost to random forests, using DFMs, TF-IDFs, shingles, etc. It’s certainly not sufficient data to train a reliable model, since we either use each line of lyrics and end up with a lot of very uninformative lines, or use each song’s lyrics and end up with just a few hundred training sets.
But, we’re here for the fun of it, and a tensorflow model with six layers using word embeddings worked not only very fast but also reached almost 80% accuracy on the unseen test set.
For example, if we feed the model the sentence “I am so happy because today I found my friends” it says it sounds very much like Sonic Youth. No surprises here.
And because tensorflow models are super easy to put into production, I built a little shiny app that allows you to check whether your own lyrics (or poems or thoughts or confessions) sound more like Swans or more like Sonic Youth.